※※※※※※※※※※※※※※※※※※※※※※※※※
※ ※
※ 行動非同步傳輸模式網路
※
※ 之行動與資源管理研究 ※
※
※
※※※※※※※※※※※※※※※※※※※※※※※※※
計畫類別：n個別型計畫 □整合型計畫
計畫編號：NSC 89 – 2213 – E – 110 – 022
執行期間： 88 年 8 月 1 日至 89 年 7 月 31 日
計畫主持人： 陳嘉玫
本成果報告包括以下應繳交之附件：
□赴國外出差或研習心得報告一份
□赴大陸地區出差或研習心得報告一份
□出席國際學術會議心得報告及發表之論文各一份
□國際合作研究計畫國外研究報告書一份
執行單位：中山大學資訊管理系
中 華 民 國 89 年 7 月 31 日
行政院國家科學委員會專題研究計畫成果報告
行動非同步傳輸模式網路
之行動與資源管理研究
Mobility and Resour ce Management
in Mobile ATM Networ ks
計畫編號：NSC 89 – 2213 – E – 110 – 022
執行期限：88 年 8 月 1 日至 89 年 7 月 31 日
主持人：陳嘉玫 中山大學資訊管理系
Abstract
This paper considers the wireless access (WASW) switch planning problem of wireless/mobile ATM networks and suggests a genetic algorithm for solving it. We first introduce our Genetic Optimization Process. We also develop analytical models to determine the tracking and locating costs for mobile switches. The fitness function is formed by these analytical models. Load balance and call-to-mobility ratio are also considered in our fitness function. Finally, we give an example scenario to demonstrate the functionality of our genetic algorithm.
Keywords: resource management, mobile
networks
1. Intr oduction
In this paper, we address the wireless access switch assignment problem to support mobile switches in an ATM network. Our Genetic Algorithm serves for finding the solution of the WASW assignment problem, i.e. a set of ATM switches on which the network designer should place the wireless communication equipment. Based on the Private Network-Network Interface (PNNI) [1] protocol, we develop analytical models to determine the tracking and locating costs. These costs are evaluated as the fitness function of our genetic optimization process. By using the genetic optimization process, we are able to obtain the WASW assignments of an ATM network according to different call-to-mobility ratios of mobile switches and different
mobility distributions of mobile switches.
2. Genetic Optimization Pr ocess
The study focuses on how to propose a method that helps network planners to obtain an optimal deployment of wireless access switches. We take the Genetic Algorithm as our search approach. In our proposed Genetic Optimization Process, a feasible solution is an assignment of WASWs on the given PNNI network. An analytical model to determine the fitness of a WASW assignment is also provided as the fitness function in the Genetic Optimization Process. The fitness function consists of two major components, namely, tracking cost and locating cost. Besides, the mobile switch geographic density distribution, call-to-mobility ratio, and load balance in our models are also to be taken account.
2.1 Functional Descr iption
The first step is to encode a WASW assignment to chromosome style. The Genetic Algorithm starts with the initial population. Offspring is then produced as the result of crossover between two randomly selected chromosomes from the population. The offspring mutates with a certain value of probability. All chromosomes are evaluated by fitness function, and then selection is proceeded. The greater fitness value a chromosome reveal, the higher opportunity it is selected. Once selection process is performed, a new generation of population is created and is ready for reproduction in the next iteration. These procedures
The Genetic Optimization Process is an iterative process that intends to find out an optimal solution among all candidate solutions. This problem consists of finding a WASW assignment with minimal cost Φ’ = min{Φ} where ) ( 2 Locating Tracking C C ρ σ + = Φ andF =−Φ (1)
Function F is the fitness function in the Genetic Optimization Process. After introducing and analyzing the cost models, the fitness function F will be explained integrally.
2.2.1 Vor onoi Diagr am
Assume that the entire network is deployed on a bounded geographic plane. When a mobile switch moves around the plane, assume that the mobile switch always connects to the "nearest" WASW. Under these assumptions, this geographic plane can be partitioned into regions, i.e., Voronoi polygons, which we denote by V(i), for a given WASW assignment. Assuming that a geographic density function f(x,y) is provided, which represents the average number of mobile switches stay at point (x, y). Given w WASWs assigned to fixed switches (s1,s2, ..., sw) on the network, they will partition the plane into w Voronoi polygons. The average number of mobile switches connect to WASW si, which we denote by αi, is the summation of density values of points within V(i), i.e. ∫∫ = (x,y)∈V(i) ( , ) i f xydydx α
.
(2) 2.2.2 Tr acking CostWe assume that no two mobile switches are directly connected together. WASWs are bridges between mobile switches and fixed network, mobile switches connect to fixed network through WASWs (wireless access switches) with wireless connection. We assume the address summarization of normal non-mobile switches is the same as described in PNNI standard. Once a mobile switch powers on or moves, it triggers a topology update process informing the related switches that topology database must be modified to reflect topology change.
When a mobile switch powers on or moves, mobile
our previous study [2].
2.2.3 Locating Cost
The PNNI 1.0 standard does not explicitly consider locating procedure for mobile switches or mobile endpoints before connection setup. Instead, connection setup proceeds with every switch "believing" its reachability information and topology database. Therefore, we propose a routing algorithm. The source switch selects a path to the destination based on the network structure known to it. Such a path is not a complete route because it does not contain the details outside the source’s peer group, containing a complete path within the source’s peer group and a sequence of LGNs outside the source’s peer group.
The locating cost is defined as the cost of setting up a connection from sender to receiver. The locating cost consists of (1) the shortest-path decision cost by the source switch and entry nodes in the PNNI hierarchy and (2) the setup cost by the switches along the path. Detail analysis also refers to our previous study [2].
2.3 Encoding Mechanism
Binary encoding is applied for assigning WASW on the topology of the lowest level in PNNI hierarchy. Let the number of fixed switches in the topology is n, and the number of WASW to be assigned is w. Each fixed switch is given a sequence number which represents where this switch is located on the chromosome. A chromosome is a bit string of size n. Every bit serves as one fixed switch on the topology. A bit is set if wireless access equipment has been facilitated on this switch.
2.4 Pr ocess Initialization
In this step, a population of size Np/2 is created so that the condition for reproduction process is well prepared. After recombination, the size of population grows up to Np and half of them will be selected for the next reproduction process. The Np-sized population is said to be one generation.
This process is initialized by creating Np/2 bit strings. Each string has w randomly selected set bits. Every bit string is distinct.
2.5 Fitness Function
In our Genetic Optimization Process, we consider not only the tracking cost and locating cost we described above, but also the load balance of each WASW assigned. For a given WASW assignment, the
variance σ2 is defined by ∑ − − = = w i i w 1 2 1 ) (α α σ (3)
where α=∑wi=1αi w. The variance 2
σ can be used to represent the load balance level of a WASW assignment. Finally, we integrate the tracking cost, locating cost, and load balance level to form our fitness function F given in equation 1.
2.6 Recombination and Selection Pr ocesses
Recombination and selection are the main part of the Genetic Algorithm. Recombination process includes crossover and mutation. After that, selection will be proceeded. The better a chromosome behaves, the more chance it will be selected. With continual operations of recombination and selection, the optimal chromosome will eventually bubble up.
2.6.1 Recombination
An example of crossover is shown in Figure 1. In this example, parents graph A and graph B crossover and make children graph A' and graph B'. Figure 2 illustrates mutation process that performed on graph C, and graph C' is the result of mutation. Beyond that, not every recombination occurs with mutation. In our Genetic Optimization Process, we give a constant probability for mutation at every recombination process.
2.6.2 Selection
Selection process determines individuals of the next generation population. In the WASW assignment problem, the greater fitness function value a chromosome evaluated, the better it behaves. The fitness function is described in section. 2.5. In our Genetic Optimization Process, Elitism Selection is adapted. Elitism Selection is quite a steady way to converge the optimal solution. There is, however, a menace to be trapped by local minimum. On the purpose of preventing that, more than once of
processes is necessary.
Figure 1. An Example of Crossover
Figure 2. An Example of Mutation
3. Example and Results
In order to visualize the optimization process by applying the Genetic Optimization Process, an example is presented. We test two kinds of density distribution and three call-to-mobility ratios.
Experiment parameters are summarized in Table 1.
Total number of switches 38
Total number of WASW 10
link short db message c c c c : : : 1:1:1:1 call-to-mobility ratio(CMR) 0.1, 1, 10 Table 1. Experiment parameters
In this example, an arbitrary network topology placed on a square plane is used This square plane is divided equally into 50 by 50 grids. A switch will be located in one of the 2500 grids. This network topology is grouped to construct a 3-level PNNI hierarchy. Further more, a density distribution which represent the average long-term distributions of mobile switch are shown in Figure 3. For one density distribution and one CMR value, we apply the Genetic Optimization Process 10 times. One Genetic Optimization Process stops at the 300th generation. We
minimal cost in these 10 candidate results. All optimal solutions for different CMR values are presented in Figure 4.
Figure 3. Density Distribution
(a) (b) (c)
Figure 4. WASW assignment: (a)ρ=0.1; (b)ρ =1; (c)ρ=10
4. Conclusions
In our study, we have solved the WASW assignment problem by Genetic Algorithm (GA). We developed a Genetic Optimization Process that can help network planners solving the WASW assignment problem. Each WASW assignment will be evaluated by a fitness function to determine its fitness. We define a function Φ consist of two costs: (1) tracking cost and (2) locating cost as the fitness function we use in the Genetic Optimization Process. In addition, we also consider the load balance and call-to-mobility ratio issue of WASW assignment in the fitness
the higher density an area is, the more WASWs flock. This result follows our intuition.
Refer ences
[1] The ATM Forum, ``ATM Private Network-to-Network Interface (PNNI) Specification Version 1.0", af-pnni-0055_0000, 1995.
[2] Sheng-Tzong Cheng, C.M. Chen, and J.W. Lin, ``Assignment Techniques for Wireless Access ATM Switches Using Genetic Algorithms”,
